Microscopic control and optimization of traffic network with Q-Learning algorithm

Chin, Yit Kwong (2013) Microscopic control and optimization of traffic network with Q-Learning algorithm. Masters thesis, Universiti Malaysia Sabah.

	Text 24 PAGES.pdf Download (823kB)
	Text FULLTEXT.pdf Restricted to Registered users only Download (11MB)

Abstract

The main objective of this research is to optimize the traffic flow within a traffic network using microscopic level control. With the increment of on-road vehicles, modern traffic networks have more complicated topologies dealing with the higher traffic demands. Fixed-time traffic signal timing plan (FTSTP) management system and Webster's model are the most common practice used by Public Works Department of Malaysia (JKR). The current efforts made by JKR should be improved to meet the increasing traffic demands within the traffic network. In this work, Q-learning traffic signal timing plan management system (QLTSTP) is developed with the attributes of Q-learning (QL). QL is able to search for the best traffic signal timing plan through the learning process from the past experiences. In order to test the viability of the proposed system, level of vehicle in queue has been chosen as the performance indicator, whilst level of traffic flow is used to determine the traffic condition of the traffic network. QLTSTP is designed as a multi-agent system with every intersection in the developed traffic network is managed by an individual QLTSTP agent towards a mutual objective. The individual QLTSTP agent has shown better performance against the FTSTP management system with the improvement of 2.61 % - 10.40% under various traffic conditions at single intersection traffic model. The performance of the multi-QLTSTP agent system is then tested in different topologies of traffic networks. The results show that there are about 2.90 % - 18.99 % of improvements in the total vehicles pass through every intersection in the traffic network for the proposed multi-QLTSP agent system. Therefore, it can be concluded that under different traffic conditions of traffic network, the proposed multi-QLTSTP agent system is able to optimize the traffic flow with the developed traffic model.The main objective of this research is to optimize the traffic flow within a traffic network using microscopic level control. With the increment of on-road vehicles, modern traffic networks have more complicated topologies dealing with the higher traffic demands. Fixed-time traffic signal timing plan (FTSTP) management system and Webster's model are the most common practice used by Public Works Department of Malaysia (JKR). The current efforts made by JKR should be improved to meet the increasing traffic demands within the traffic network. In this work, Q-learning traffic signal timing plan management system (QLTSTP) is developed with the attributes of Q-learning (QL). QL is able to search for the best traffic signal timing plan through the learning process from the past experiences. In order to test the viability of the proposed system, level of vehicle in queue has been chosen as the performance indicator, whilst level of traffic flow is used to determine the traffic condition of the traffic network. QLTSTP is designed as a multi-agent system with every intersection in the developed traffic network is managed by an individual QLTSTP agent towards a mutual objective. The individual QLTSTP agent has shown better performance against the FTSTP management system with the improvement of 2.61 % - 10.40% under various traffic conditions at single intersection traffic model. The performance of the multi-QLTSTP agent system is then tested in different topologies of traffic networks. The results show that there are about 2.90 % - 18.99 % of improvements in the total vehicles pass through every intersection in the traffic network for the proposed multi-QLTSP agent system. Therefore, it can be concluded that under different traffic conditions of traffic network, the proposed multi-QLTSTP agent system is able to optimize the traffic flow with the developed traffic model.

Item Type:	Thesis (Masters)
Keyword:	Traffic flow, Traffic network, Microscopic control, Fixed-time traffic signal timing plan
Subjects:	T Technology > TE Highway engineering. Roads and pavements > TE1-450 Highway engineering. Roads and pavements > TE210-228.3 Construction details Including foundations, maintenance, equipment
Department:	SCHOOL > School of Engineering and Information Technology
Depositing User:	DG MASNIAH AHMAD -
Date Deposited:	09 Feb 2024 11:10
Last Modified:	09 Feb 2024 11:10
URI:	https://eprints.ums.edu.my/id/eprint/38168

Actions (login required)

View Item